.png){kind=icon}
.png){kind=icon}
Return
Person Re-Identification With Multi-Features Based on Evolutionary Algorithm
.png){kind=icon}
PRE
AI
DOI：10.1109/TETCI.2021.3122995.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Person re-identification can identify a particular pedestrian automatically by computer in surveillance video, who has occurred in the monitoring network. It is one of the hot topics in the field of intelligent video surveillance. However, it faces many challenges. For example, robust feature representation model and ideal similarity measure method are both lacking, which will seriously affect the identification accuracy. These problems can be solved to some extent through applying appropriate optimization algorithm, which could optimize feature selection process, and guide similarity measure method design. Evolutionary algorithm is a good optimization method, but two obvious drawbacks have been found in practical application, which are slow convergence speed and easy convergence to local optimal solution respectively. Therefore, traditional evolutionary algorithm is improved in this paper firstly. The existing problem will be solved by improving search efficiency and maintaining population diversity, after that, improved evolutionary algorithm is applied to optimize the process of person re-identification. Feature representation model with high robustness will be designed, and reasonable and effective similarity measure method will be obtained through learning. Thus, the identification accuracy and system performance will be improved. The algorithm proposed in this paper has been compared with some classical algorithms and the case without using evolutionary algorithm to optimize. Experimental results show that method in this paper is an effective person re-identification scheme.
Keywords:
Evolutionary computation
Feature extraction
Image color analysis
Statistics
Sociology
Convergence
Search problems
Person Re-identification
Evolutionary Algorithm
Feature Representation
Feature Matching
Journal
I
IF:
6.5
Papers:
1.4K
Citations:
4.5K